Papers.

Birds of a Feather Tweet Together: Integrating Network and Content Analyses to Examine Cross-Ideology Exposure on Twitter

Building a privacy-preserving semantic overlay for Peer-to-Peer networks

A Method for Discovering Dynamic Network Motifs by Encoding Topic Propagation

Understanding the role of constraints on architecturally significant requirements

A key constraint on software development is reliance on tools, which we define as COTS products, software services, languages, frameworks and platforms. These tools may have significant architectural impacts that are not obvious when the requirements are elicited, tools selected, and architecture sketched out. In this paper, we report on a case study we conducted to identify architecturally significant requirements (ASRs) that were impacted by tool selection. We identified ASRs in an existing health IT project, CONNECT, and also identified the constraints on the project that were tool-related. We produce a mapping showing how the architectural risks identified in the initial architectural analysis were impacted by the tool choices made. We produce metrics showing how much time has been consumed when implementing ASRs that involve working around/with these constraints and the risks associated with them.

Application Portability in Cloud Computing: An Abstraction-Driven Perspective

Cloud computing has changed the way organizations create, manage, and evolve their applications. While the abundance of computing resources at low cost opens up many possibilities for migrating applications to the cloud, this migration also comes at a price. Cloud applications, in many cases, depend on certain provider specific features or services. In moving applications to the cloud, application developers face the challenge of balancing these dependencies to avoid vendor lock-in. We present an abstraction-driven approach to address the application portability issues and focus on the application development process. We also present our theoretical basis and experience in two practical projects where we have applied the abstraction-driven approach.

Systemic Risk and User-Level Performance in Private P2P Communities

Learning Hierarchical Features for Scene Labeling

Scene labeling consists in labeling each pixel in an image with the category of the object it belongs to. We propose a method that uses a multiscale convolutional network trained from raw pixels to extract dense feature vectors that encode regions of multiple sizes centered on each pixel. The method alleviates the need for engineered features, and produces a powerful representation that captures texture, shape and contextual information. We report results using multiple post-processing methods to produce the final labeling. Among those, we propose a technique to automatically retrieve, from a pool of segmentation components, an optimal set of components that best explain the scene; these components are arbitrary, e.g. they can be taken from a segmentation tree, or from any family of over-segmentations. The system yields record accuracies on the Sift Flow Dataset (33 classes) and the Barcelona Dataset (170 classes) and near-record accuracy on Stanford Background Dataset (8 classes), while being an order of magnitude faster than competing approaches, producing a 320 × 240 image labeling in less than a second, including feature extraction.

A Convex Formulation for Learning a Shared Predictive Structure from Multiple Tasks

Electrochemical Model Based Observer Design for a Lithium-Ion Battery

Database-Driven Real-Time Heuristic Search in Video-Game Pathfinding.

An Ontology-Based Framework for Model-Driven Analysis of Situations in Data Centers

Investment Strategies for Credit-Based P2P Communities

Leveraging node properties in random walks for robust reputations in decentralized networks

Through the wormhole: Low cost, fresh peer sampling for the Internet

Application of information diffusion in social networks: Verifying the use of web/blog topology entropy as an indicator for real world impact

Using GPUs to Accelerate CAD Algorithms

Sigma hulls for Gaussian belief space planning for imprecise articulated robots amid obstacles

In many home and service applications, an emerging class of articulated robots such as the Raven and Baxter trade off precision in actuation and sensing to reduce costs and to reduce the potential for injury to humans in their workspaces. For planning and control of such robots, planning in belief ssigma hullpace, i.e., modeling such problems as POMDPs, has shown great promise but existing belief space planning methods have primarily been applied to cases where robots can be approximated as points or spheres. In this paper, we extend the belief space framework to treat articulated robots where the linkage can be decomposed into convex components. To allow planning and collision avoidance in Gaussian belief spaces, we introduce the concept of sigma hulls: convex hulls of robot links transformed according to the sigma standard deviation boundary points generated by the Unscented Kalman filter (UKF). We characterize the signed distances between sigma hulls and obstacles in the workspace to formulate efficient collision avoidance constraints compatible with the Gilbert-Johnson-Keerthi (GKJ) and Expanding Polytope Algorithms (EPA) within an optimization-based planning framework. We report results in simulation for planning motions for a 4-DOF planar robot and a 7-DOF articulated robot with imprecise actuation and inaccurate sensors. These experiments suggest that the sigma hull framework can significantly reduce the probability of collision and is computationally efficient enough to permit iterative re-planning for model predictive control.

The FUSE domain-aware approach to user model interoperability: A comparative study

Constrained by limited information and recommendation opportunities: An exploration and exploitation problem for recommender systems

Understanding user behavior in Spotify

Crowdsourcing GUI Tests

Variations on Using Propagation Cost to Measure Architecture Modifiability Properties

Tools available for measuring the modifiability or impact of change of a system through its architecture typically use structural metrics. These metrics take into account dependencies among the different elements of a system. However, they fail to capture the semantics of an architectural transformation necessary to control the complexity and cost of making changes. To highlight such limitations, this paper presents a study where we applied a representative structural metric, called 'propagation cost', to archetypical architectural transformations known to affect system modifiability such as rearchitecting a tightly coupled system to a layered pattern. We observe that in its original form the propagation cost metric does not provide consistent indications of architecture health. Enhancing this metric based on the semantics of the architectural pattern and tactics used in the transformation show improvements. Our results demonstrate that these enhancements detect modifiability properties that are not detectable by the propagation cost metric.

Kinodynamic RRT*: Asymptotically optimal motion planning for robots with linear dynamics

We present Kinodynamic RRT*, an incremental sampling-based approach for asymptotically optimal motion planning for robots with linear dynamics. Our approach extends RRT*, which was introduced for holonomic robots [10], by using a fixed-final-state-free-final-time controller that optimally connects any pair of states, where the cost function is expressed as a trade-off between the duration of a trajectory and the expended control effort. Our approach generalizes earlier work on RRT* for kinodynamic systems, as it guarantees asymptotic optimality for any system with controllable linear dynamics, in state spaces of any dimension. In addition, we show that for the rich subclass of systems with a nilpotent dynamics matrix, closed-form solutions for optimal trajectories can be derived, which keeps the computational overhead of our algorithm compared to traditional RRT* at a minimum. We demonstrate the potential of our approach by computing asymptotically optimal trajectories in three challenging motion planning scenarios: (i) a planar robot with a 4-D state space and double integrator dynamics, (ii) an aerial vehicle with a 10-D state space and linearized quadrotor dynamics, and (iii) a car-like robot with a 5-D state space and non-linear dynamics.

Reciprocal collision avoidance for robots with linear dynamics using LQR-Obstacles

In this paper we present a formal approach to reciprocal collision avoidance for multiple mobile robots sharing a common 2-D or 3-D workspace whose dynamics are subject to linear differential constraints. Our approach defines a protocol for robots to select their control input independently (i.e. without coordination with other robots) while guaranteeing collision-free motion for all robots, assuming the robots can perfectly observe each other's state. To this end, we extend the concept of LQR-Obstacles (which is a generalization of Velocity Obstacles to robots with dynamics for collision avoidance among static obstacles) for reciprocal collision avoidance among multiple robots. We implemented and tested our approach in 3-D simulation environments for reciprocal collision avoidance of quadrotor helicopters, which have complex dynamics in 16-D state spaces. Our results show that our approach enables collision avoidance among over a hundred quadrotors in tight workspaces at real-time computation rates.

Causal graph-based video segmentation

Numerous approaches in image processing and computer vision are making use of super-pixels as a pre-processing step. Among the different methods producing such over-segmentation of an image, the graph-based approach of Felzenszwalb and Huttenlocher is broadly employed. One of its interesting properties is that the regions are computed in a greedy manner in quasi-linear time. The algorithm may be trivially extended to video segmentation by considering a video as a 3D volume, however, this can not be the case for causal segmentation, when subsequent frames are unknown. We propose an efficient video segmentation approach that computes temporally consistent pixels in a causal manner, filling the need for causal and real time applications.

MLI: An API for Distributed Machine Learning

Performance optimization for distributed intra-node-parallel streaming systems

Aeolus: An optimizer for distributed intra-node-parallel streaming systems

Lightweight gossip-based distribution estimation

Using multiple versions of speech input in phone recognition

Articulatory trajectories for large-vocabulary speech recognition

Scale-space expansion of acoustic features improves speech event detection

Hardware-accelerated regular expression matching for high-throughput text analytics

Automatically captured sociability and sleep quality in healthy adults

NodeXL: Simple network analysis for social media

Real Time Self-Organizing Systems for Swarm Control

AptStore: Dynamic storage management for hadoop

Keynote and invited speakers abstracts

Tracking with deep neural networks

We present deep neural network models applied to tracking objects of interest. Deep neural networks trained for general-purpose use are introduced to conduct long-term tracking, which requires scale-invariant feature extraction even when the object dramatically changes shape as it moves in the scene. We use two-layer networks trained using either supervised or unsupervised learning techniques. The networks, augmented with a radial basis function classifier, are able to track objects based on a single example. We tested the networks tracking capability on the TLD dataset, one of the most difficult sets of tracking tasks and real-time tracking is achieved in 0.074 seconds per frame for 320×240 pixel image on a 2-core 2.7GHz Intel i7 laptop.

Time series compression for adaptive chart generation.

Continuous hyperparameter optimization for large-scale recommender systems

Measuring inter-site engagement

Invited Abstract: Ricardo Baez-Yates

Measuring the $t \bar th$ coupling from same-sign dilepton $+2b$ measurements

The Population of Giant Clumps in Simulated High-z Galaxies: In-situ and Ex-situ, Migration and Survival

We study the properties of giant clumps and their radial gradients in high-$z$ disc galaxies using AMR cosmological simulations. Our sample consists of 770 snapshots in the redshift range $z=4-1$ from 29 galaxies that at $z=2$ span the stellar mass range $(0.2-3)\times 10^{11}M_{\odot}$. Extended gas discs exist in 83% of the snapshots. Clumps are identified by gas density in 3D and their stellar and dark matter components are considered thereafter. While most of the overdensities are diffuse and elongated, 91% of their mass and 83% of their star-fromation rate (SFR) are in compact round clumps. Nearly all galaxies have a central, massive bulge clump, while 70% of the discs show off-center clumps, 3-4 per galaxy. The fraction of clumpy discs peaks at intermediate disc masses. Clumps are divided based on dark-matter content into $\textit{in-situ}$ and $\textit{ex-situ}$, originating from violent disc instability (VDI) and minor mergers respectively. 60% of the discs are in a VDI phase showing off-center $\textit{in-situ}$ clumps, which contribute 1-7% of the disc mass and 5-45% of its SFR. The $\textit{in-situ}$ clumps constitute 75% of the off-center clumps in terms of number and SFR but only half the mass, each clump containing on average 1% of the disc mass and 6% of its SFR. They have young stellar ages, $100-400 {\rm Myr}$, and high specific SFR (sSFR), $1-10 {\rm Gyr}^{-1}$. They exhibit gradients resulting from inward clump migration, where the inner clumps are somewhat more massive and older, with lower gas fraction and sSFR and higher metallicity. Similar observed gradients indicate that clumps survive outflows. The \exsitu clumps have stellar ages $0.5-3 {\rm Gyr}$ and sSFR $\sim 0.1-2 {\rm Gyr}^{-1}$, and they exhibit weaker gradients. Massive clumps of old stars at large radii are likely \exsitu mergers, though half of them share the disc rotation.

Benchmarking Big Data Systems and the BigData Top100 List

The AGORA High-Resolution Galaxy Simulations Comparison Project

We introduce the AGORA project, a comprehensive numerical study of well-resolved galaxies within the LCDM cosmology. Cosmological hydrodynamic simulations with force resolutions of ~100 proper pc or better will be run with a variety of code platforms to follow the hierarchical growth, star formation history, morphological transformation, and the cycle of baryons in and out of 8 galaxies with halo masses M_vir ~= 1e10, 1e11, 1e12, and 1e13 Msun at z=0 and two different ("violent" and "quiescent") assembly histories. The numerical techniques and implementations used in this project include the smoothed particle hydrodynamics codes GADGET and GASOLINE, and the adaptive mesh refinement codes ART, ENZO, and RAMSES. The codes will share common initial conditions and common astrophysics packages including UV background, metal-dependent radiative cooling, metal and energy yields of supernovae, and stellar initial mass function. These are described in detail in the present paper. Subgrid star formation and feedback prescriptions will be tuned to provide a realistic interstellar and circumgalactic medium using a non-cosmological disk galaxy simulation. Cosmological runs will be systematically compared with each other using a common analysis toolkit, and validated against observations to verify that the solutions are robust - i.e., that the astrophysical assumptions are responsible for any success, rather than artifacts of particular implementations. The goals of the AGORA project are, broadly speaking, to raise the realism and predictive power of galaxy simulations and the understanding of the feedback processes that regulate galaxy "metabolism." The proof-of-concept dark matter-only test of the formation of a galactic halo with a z=0 mass of M_vir ~= 1.7e11 Msun by 9 different versions of the participating codes is also presented to validate the infrastructure of the project.

A Comparison between Semi-Analytic Model Predictions for the CANDELS Survey

We compare the predictions of three independently developed semi-analytic galaxy formation models that are being used to aid in the interpretation of results from the CANDELS survey. These models are each applied to the same set of halo merger trees extracted from the "Bolshoi" simulation and are carefully tuned to match the local galaxy stellar mass function using the powerful method of Bayesian Inference coupled with MCMC or by hand. The comparisons reveal that in spite of the significantly different parameterizations for star formation and feedback processes, the three models yield qualitatively similar predictions for the assembly histories of galaxy stellar mass and star formation over cosmic time. We show that the SAMs generally require strong outflows to suppress star formation in low-mass halos to match the present day stellar mass function. However, all of the models considered produce predictions for the star formation rates and metallicities of low-mass galaxies that are inconsistent with existing data and diverge between the models. We suggest that large differences in the metallicity relations and small differences in the stellar mass assembly histories of model galaxies stem from different assumptions for the outflow mass-loading factor. Importantly, while more accurate observational measurements for stellar mass, SFR and metallicity of galaxies at 1<z<5 will discriminate between models, the discrepancies between the models and existing data of these observables have already revealed challenging problems in understanding star formation and its feedback in galaxy formation. The three sets of models are being used to construct catalogs of mock galaxies on light cones that have the same geometry as the CANDELS survey, which should be particularly useful for quantifying the biases and uncertainties on measurements and inferences from the real observations. -ABRIDGED

CANDELS+3D-HST: compact SFGs at z~2-3, the progenitors of the first quiescent galaxies

We analyze the star-forming and structural properties of 45 massive (log(M/Msun)>10) compact star-forming galaxies (SFGs) at 2<z<3 to explore whether they are progenitors of compact quiescent galaxies at z~2. The optical/NIR and far-IR Spitzer/Herschel colors indicate that most compact SFGs are heavily obscured. Nearly half (47%) host an X-ray bright AGN. In contrast, only about 10% of other massive galaxies at that time host AGNs. Compact SFGs have centrally-concentrated light profiles and spheroidal morphologies similar to quiescent galaxies, and are thus strikingly different from other SFGs. Most compact SFGs lie either within the SFR-M main sequence (65%) or below (30%), on the expected evolutionary path towards quiescent galaxies. These results show conclusively that galaxies become more compact before they lose their gas and dust, quenching star formation. Using extensive HST photometry from CANDELS and grism spectroscopy from the 3D-HST survey, we model their stellar populations with either exponentially declining (tau) star formation histories (SFHs) or physically-motivated SFHs drawn from semi-analytic models (SAMs). SAMs predict longer formation timescales and older ages ~2 Gyr, which are nearly twice as old as the estimates of the tau models. While both models yield good SED fits, SAM SFHs better match the observed slope and zero point of the SFR-M main sequence. Some low-mass compact SFGs (log(M/Msun)=10-10.6) have younger ages but lower sSFRs than that of more massive galaxies, suggesting that the low-mass galaxies reach the red sequence faster. If the progenitors of compact SFGs are extended SFGs, state-of-the-art SAMs show that mergers and disk instabilities are both able to shrink galaxies, but disk instabilities are more frequent (60% versus 40%) and form more concentrated galaxies. We confirm this result via high-resolution hydrodynamic simulations.

Geodesic Universality in General Relativity